Thomas Lingner

BCI competition 2003-data set IIb: support vector machines for the P300 speller paradigm

We propose an approach to analyze data from the P300 speller paradigm using the machine-learning technique support vector machines. In a conservative classification scheme, we found the correct solution after five repetitions. While the classification within the competition is designed for offline analysis, our approach is also well-suited for a real-world online solution: It is fast, requires only 10 electrode positions and demands only a small amount of preprocessing.

Orphelia: predicting genes in metagenomic sequencing reads

Metagenomic sequencing projects yield numerous sequencing reads of a diverse range of uncultivated and mostly yet unknown microorganisms. In many cases, these sequencing reads cannot be assembled into longer contigs. Thus, gene prediction tools that were originally developed for whole-genome analysis are not suitable for processing metagenomes. Orphelia is a program for predicting genes in short DNA sequences that is available through a web server application (http://orphelia.gobics.de). Orphelia utilizes prediction models that were created with machine learning techniques on the basis of a wide range of annotated genomes. In contrast to other methods for metagenomic gene prediction, Orphelia has fragment length-specific prediction models for the two most popular sequencing techniques in metagenomics, chain termination sequencing and pyrosequencing. These models ensure highly specific gene predictions.

Identification of Novel Plant Peroxisomal Targeting Signals by a Combination of Machine Learning Methods and in Vivo Subcellular Targeting Analyses

Two different prediction methods for Arabidopsis proteins carrying peroxisome targeting signals type 1 (PTS1) are described. Validation of many novel targeting signals and Arabidopsis PTS1 proteins by in vivo localization experiments demonstrates a high prediction accuracy of the new methods. In the postgenomic era, accurate prediction tools are essential for identification of the proteomes of cell organelles. Prediction methods have been developed for peroxisome-targeted proteins in animals and fungi but are missing specifically for plants. For development of a predictor for plant proteins carrying peroxisome targeting signals type 1 (PTS1), we assembled more than 2500 homologous plant sequences, mainly from EST databases. We applied a discriminative machine learning approach to derive two different prediction methods, both of which showed high prediction accuracy and recognized specific targeting-enhancing patterns in the regions upstream of the PTS1 tripeptides. Upon application of these methods to the Arabidopsis thaliana genome, 392 gene models were predicted to be peroxisome targeted. These predictions were extensively tested in vivo, resulting in a high experimental verification rate of Arabidopsis proteins previously not known to be peroxisomal. The prediction methods were able to correctly infer novel PTS1 tripeptides, which even included novel residues. Twenty-three newly predicted PTS1 tripeptides were experimentally confirmed, and a high variability of the plant PTS1 motif was discovered. These prediction methods will be instrumental in identifying low-abundance and stress-inducible peroxisomal proteins and defining the entire peroxisomal proteome of Arabidopsis and agronomically important crop plants.

Gene prediction in metagenomic fragments: A large scale machine learning approach

BackgroundMetagenomics is an approach to the characterization of microbial genomes via the direct isolation of genomic sequences from the environment without prior cultivation. The amount of metagenomic sequence data is growing fast while computational methods for metagenome analysis are still in their infancy. In contrast to genomic sequences of single species, which can usually be assembled and analyzed by many available methods, a large proportion of metagenome data remains as unassembled anonymous sequencing reads. One of the aims of all metagenomic sequencing projects is the identification of novel genes. Short length, for example, Sanger sequencing yields on average 700 bp fragments, and unknown phylogenetic origin of most fragments require approaches to gene prediction that are different from the currently available methods for genomes of single species. In particular, the large size of metagenomic samples requires fast and accurate methods with small numbers of false positive predictions.ResultsWe introduce a novel gene prediction algorithm for metagenomic fragments based on a two-stage machine learning approach. In the first stage, we use linear discriminants for monocodon usage, dicodon usage and translation initiation sites to extract features from DNA sequences. In the second stage, an artificial neural network combines these features with open reading frame length and fragment GC-content to compute the probability that this open reading frame encodes a protein. This probability is used for the classification and scoring of gene candidates. With large scale training, our method provides fast single fragment predictions with good sensitivity and specificity on artificially fragmented genomic DNA. Additionally, this method is able to predict translation initiation sites accurately and distinguishes complete from incomplete genes with high reliability.ConclusionLarge scale machine learning methods are well-suited for gene prediction in metagenomic DNA fragments. In particular, the combination of linear discriminants and neural networks is promising and should be considered for integration into metagenomic analysis pipelines. The data sets can be downloaded from the URL provided (see Availability and requirements section).

Mixture models for analysis of the taxonomic composition of metagenomes

Motivation: Inferring the taxonomic profile of a microbial community from a large collection of anonymous DNA sequencing reads is a challenging task in metagenomics. Because existing methods for taxonomic profiling of metagenomes are all based on the assignment of fragmentary sequences to phylogenetic categories, the accuracy of results largely depends on fragment length. This dependence complicates comparative analysis of data originating from different sequencing platforms or resulting from different preprocessing pipelines. Results: We here introduce a new method for taxonomic profiling based on mixture modeling of the overall oligonucleotide distribution of a sample. Our results indicate that the mixture-based profiles compare well with taxonomic profiles obtained with other methods. However, in contrast to the existing methods, our approach shows a nearly constant profiling accuracy across all kinds of read lengths and it operates at an unrivaled speed. Availability: A platform-independent implementation of the mixture modeling approach is available in terms of a MATLAB/Octave toolbox at http://gobics.de/peter/taxy. In addition, a prototypical implementation within an easy-to-use interactive tool for Windows can be downloaded. Contact: pmeinic@gwdg.de; thomas@gobics.de Supplementary Information: Supplementary data are available at Bioinformatics online.

Remote homology detection based on oligomer distances

MOTIVATION Remote homology detection is among the most intensively researched problems in bioinformatics. Currently discriminative approaches, especially kernel-based methods, provide the most accurate results. However, kernel methods also show several drawbacks: in many cases prediction of new sequences is computationally expensive, often kernels lack an interpretable model for analysis of characteristic sequence features, and finally most approaches make use of so-called hyperparameters which complicate the application of methods across different datasets. RESULTS We introduce a feature vector representation for protein sequences based on distances between short oligomers. The corresponding feature space arises from distance histograms for any possible pair of K-mers. Our distance-based approach shows important advantages in terms of computational speed while on common test data the prediction performance is highly competitive with state-of-the-art methods for protein remote homology detection. Furthermore the learnt model can easily be analyzed in terms of discriminative features and in contrast to other methods our representation does not require any tuning of kernel hyperparameters. AVAILABILITY Normalized kernel matrices for the experimental setup can be downloaded at www.gobics.de/thomas. Matlab code for computing the kernel matrices is available upon request. CONTACT thomas@gobics.de, peter@gobics.de.

Peroxisomal lactate dehydrogenase is generated by translational readthrough in mammals

Translational readthrough gives rise to low abundance proteins with C-terminal extensions beyond the stop codon. To identify functional translational readthrough, we estimated the readthrough propensity (RTP) of all stop codon contexts of the human genome by a new regression model in silico, identified a nucleotide consensus motif for high RTP by using this model, and analyzed all readthrough extensions in silico with a new predictor for peroxisomal targeting signal type 1 (PTS1). Lactate dehydrogenase B (LDHB) showed the highest combined RTP and PTS1 probability. Experimentally we show that at least 1.6% of the total cellular LDHB is targeted to the peroxisome by a conserved hidden PTS1. The readthrough-extended lactate dehydrogenase subunit LDHBx can also co-import LDHA, the other LDH subunit, into peroxisomes. Peroxisomal LDH is conserved in mammals and likely contributes to redox equivalent regeneration in peroxisomes. DOI: http://dx.doi.org/10.7554/eLife.03640.001

PredPlantPTS1: A Web Server for the Prediction of Plant Peroxisomal Proteins

Prediction of subcellular protein localization is essential to correctly assign unknown proteins to cell organelle-specific protein networks and to ultimately determine protein function. For metazoa, several computational approaches have been developed in the past decade to predict peroxisomal proteins carrying the peroxisome targeting signal type 1 (PTS1). However, plant-specific PTS1 protein prediction methods have been lacking up to now, and pre-existing methods generally were incapable of correctly predicting low-abundance plant proteins possessing non-canonical PTS1 patterns. Recently, we presented a machine learning approach that is able to predict PTS1 proteins for higher plants (spermatophytes) with high accuracy and which can correctly identify unknown targeting patterns, i.e., novel PTS1 tripeptides and tripeptide residues. Here we describe the first plant-specific web server PredPlantPTS1 for the prediction of plant PTS1 proteins using the above-mentioned underlying models. The server allows the submission of protein sequences from diverse spermatophytes and also performs well for mosses and algae. The easy-to-use web interface provides detailed output in terms of (i) the peroxisomal targeting probability of the given sequence, (ii) information whether a particular non-canonical PTS1 tripeptide has already been experimentally verified, and (iii) the prediction scores for the single C-terminal 14 amino acid residues. The latter allows identification of predicted residues that inhibit peroxisome targeting and which can be optimized using site-directed mutagenesis to raise the peroxisome targeting efficiency. The prediction server will be instrumental in identifying low-abundance and stress-inducible peroxisomal proteins and defining the entire peroxisomal proteome of Arabidopsis and agronomically important crop plants. PredPlantPTS1 is freely accessible at ppp.gobics.de.

CoMet—a web server for comparative functional profiling of metagenomes

Analyzing the functional potential of newly sequenced genomes and metagenomes has become a common task in biomedical and biological research. With the advent of high-throughput sequencing technologies comparative metagenomics opens the way to elucidate the genetically determined similarities and differences of complex microbial communities. We developed the web server ‘CoMet’ (http://comet.gobics.de), which provides an easy-to-use comparative metagenomics platform that is well-suitable for the analysis of large collections of metagenomic short read data. CoMet combines the ORF finding and subsequent assignment of protein sequences to Pfam domain families with a comparative statistical analysis. Besides comprehensive tabular data files, the CoMet server also provides visually interpretable output in terms of hierarchical clustering and multi-dimensional scaling plots and thus allows a quick overview of a given set of metagenomic samples.

Non-canonical peroxisome targeting signals: identification of novel PTS1 tripeptides and characterization of enhancer elements by computational permutation analysis

BackgroundHigh-accuracy prediction tools are essential in the post-genomic era to define organellar proteomes in their full complexity. We recently applied a discriminative machine learning approach to predict plant proteins carrying peroxisome targeting signals (PTS) type 1 from genome sequences. For Arabidopsis thaliana 392 gene models were predicted to be peroxisome-targeted. The predictions were extensively tested in vivo, resulting in a high experimental verification rate of Arabidopsis proteins previously not known to be peroxisomal.ResultsIn this study, we experimentally validated the predictions in greater depth by focusing on the most challenging Arabidopsis proteins with unknown non-canonical PTS1 tripeptides and prediction scores close to the threshold. By in vivo subcellular targeting analysis, three novel PTS1 tripeptides (QRL>, SQM>, and SDL>) and two novel tripeptide residues (Q at position −3 and D at pos. -2) were identified. To understand why, among many Arabidopsis proteins carrying the same C-terminal tripeptides, these proteins were specifically predicted as peroxisomal, the residues upstream of the PTS1 tripeptide were computationally permuted and the changes in prediction scores were analyzed. The newly identified Arabidopsis proteins were found to contain four to five amino acid residues of high predicted targeting enhancing properties at position −4 to −12 in front of the non-canonical PTS1 tripeptide. The identity of the predicted targeting enhancing residues was unexpectedly diverse, comprising besides basic residues also proline, hydroxylated (Ser, Thr), hydrophobic (Ala, Val), and even acidic residues.ConclusionsOur computational and experimental analyses demonstrate that the plant PTS1 tripeptide motif is more diverse than previously thought, including an increasing number of non-canonical sequences and allowed residues. Specific targeting enhancing elements can be predicted for particular sequences of interest and are far more diverse in amino acid composition and positioning than previously assumed. Machine learning methods become indispensable to predict which specific proteins, among numerous candidate proteins carrying the same non-canonical PTS1 tripeptide, contain sufficient enhancer elements in terms of number, positioning and total strength to cause peroxisome targeting.